Design Doc: Distributed Lookup Table Operator
A distribute lookup table operator in PaddlePaddle where the table could be out of the memory of a computer.
A lookup table operator is well-used in deep learning for learning the representation, or the embedding, of symbols.
The Forward Algorithm
The forward algorithm of the lookup table is a multiplication of the input vector x and the lookup table matrix W:
$$y = x * W$$
When x is a sparse vector of symbols, the above multiplication simplifies into looking up rows in W that correspond to symbols in x, denoted by W(x). Please be aware that W could be huge and out of the memory, so we'd need a distributed storage service, which supports the lookup of rows.
The following figure illustrates the multiplication of x with two non-zero elements, or say two symbols, and a lookup table W:
The Backward Algorithm
The backward algorithm computes W'(x) using W(x). W'(x) has the same the scale of size as W(x) and is much smaller than W.
To optimize W given W', we can do simple SGD update:
$$W = f(W') = \lambda * W'$$
or some more sophisticated algorithms that rely on both W' and W:
$$W = f(W, W')$$
The following figure illustrates the backward pass of the lookup operator:
Distributed Lookup Table
Problem 1: The lookup table may be very large.
In the condition like the search engine and recommendation system, the number of feature Id may be very large, say 100,000,000,000, then for a float value lookup table of size 8, the total size of the table is:
100,000,000,000 * 8 * 4(Bytes) = 2980.23 GB
Solution: Distributed storage
Paddle use SelectedRows as the storage format for the lookup table, the lookup table parameter will be split to multi-machine according to the hash of the feature ID, and data will also be split and send to the same machine to prefetch the parameter.
For common parameters, the trainer will get the whole parameter for training, but for the big lookup table, the trainer can not store the whole parameter. Because the input data feature is very sparse, every time we only need a few parameters for training, so we use
prefetch_opto only prefetch the parameter needed to trainer.
Problem 2. The Id in the lookup table is not sure before training.
The feature Id is calculated by the hash function because the feature data source is so large, we can not get all the Id before training. So we can not initialize the table before training.
Solution: Id auto growth
At the beginning of training, paddle only malloc the memory for the lookup table at parameter server side, the Id and it's value will not be initialized. During training, when a parameter server received an Id, if it is already in the lookup table, it will return the existing parameter, if the Id does not exist, paddle will add it into the lookup table and initialize the value for it.
Problem 3: parameter load and save
For common parameters, paddle use trainer to save and load them. But for distributed lookup table, trainer cannot do this because it's large size.
Solution: Parameter server side save and load
Paddle support parameter server side save and load for distribute lookup table. Each machine of parameter servers will only save and load part of the whole table.
The whole architecture of the distribute lookup table is as below:
- Read a batch of data, the data is feature ids.
- The input ids will be split by
split_ids_opwith the same hash function of the lookup table.
prefetch_opuse the split result to prefetch parameters back from the lookup table.
- Run forward-backward to get the gradient of the lookup table.
split_ids_opsplit the gradient and then use
send_opto the parameter server.
- parameter server update the table with the received gradient.